File size: 11,013 Bytes
e479c46 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 | # SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""CPU-only tests for the schema-fingerprint cache in ``generate_rel_stats``.
Without the fingerprint guard, an existing ``meta/relative_stats.json`` was
reused whenever a per-embodiment ``action_key`` name matched, regardless of
whether the inputs that drive the computation (``delta_indices``, ``format``,
``state_key``, ...) had since changed — silently corrupting normalization at
training time.
"""
from dataclasses import replace
import json
from unittest.mock import patch
from gr00t.configs.data.embodiment_configs import MODALITY_CONFIGS
from gr00t.data.embodiment_tags import EmbodimentTag
from gr00t.data.stats import (
LE_ROBOT_REL_STATS_FILENAME,
STATS_FINGERPRINTS_KEY,
_compute_relative_action_fingerprint,
generate_rel_stats,
)
from gr00t.data.types import ActionFormat
import numpy as np
import pytest
# Real registered embodiment with two RELATIVE action keys (eef_9d, joint_position)
# and one ABSOLUTE (gripper_position) — exercises the cache path without
# needing parquet data once calculate_stats_for_key is mocked.
EMBODIMENT = EmbodimentTag.OXE_DROID_RELATIVE_EEF_RELATIVE_JOINT
RELATIVE_KEYS = ("eef_9d", "joint_position")
def _stub_stats():
"""Synthetic stats payload shaped like calculate_stats_for_key's return."""
return {
"max": np.ones(9, dtype=np.float32),
"min": -np.ones(9, dtype=np.float32),
"q01": -np.ones(9, dtype=np.float32) * 0.99,
"q99": np.ones(9, dtype=np.float32) * 0.99,
"mean": np.zeros(9, dtype=np.float32),
"std": np.ones(9, dtype=np.float32),
}
@pytest.fixture
def dataset_dir(tmp_path):
(tmp_path / "meta").mkdir()
return tmp_path
@pytest.fixture
def mock_calculate(monkeypatch):
"""Replace the heavy parquet-driven computation with a counter-stub."""
calls = []
def fake(dataset_path, embodiment_tag, group_key, max_episodes=-1):
calls.append((str(dataset_path), embodiment_tag.value, group_key))
return _stub_stats()
monkeypatch.setattr("gr00t.data.stats.calculate_stats_for_key", fake)
return calls
# ---------------------------------------------------------------------------
# _compute_relative_action_fingerprint — pure helper, no I/O
# ---------------------------------------------------------------------------
class TestFingerprintHelper:
def test_deterministic(self):
a = _compute_relative_action_fingerprint(EMBODIMENT, "eef_9d")
b = _compute_relative_action_fingerprint(EMBODIMENT, "eef_9d")
assert a == b
def test_format_prefix(self):
fp = _compute_relative_action_fingerprint(EMBODIMENT, "eef_9d")
assert fp.startswith("sha256:") and len(fp) == len("sha256:") + 64
def test_distinct_per_action_key(self):
a = _compute_relative_action_fingerprint(EMBODIMENT, "eef_9d")
b = _compute_relative_action_fingerprint(EMBODIMENT, "joint_position")
assert a != b
@pytest.mark.parametrize(
"mutator",
[
pytest.param(
lambda cfg: replace(cfg, format=ActionFormat.XYZ_ROTVEC),
id="format-change",
),
pytest.param(
lambda cfg: replace(cfg, state_key="other_state_key"),
id="state-key-change",
),
],
)
def test_changes_when_action_config_changes(self, mutator):
"""Mutating fields the loader actually reads must invalidate the cache."""
cfgs = MODALITY_CONFIGS[EMBODIMENT.value]
action_modality = cfgs["action"]
idx = action_modality.modality_keys.index("eef_9d")
original_action_config = action_modality.action_configs[idx]
baseline = _compute_relative_action_fingerprint(EMBODIMENT, "eef_9d")
new_configs = list(action_modality.action_configs)
new_configs[idx] = mutator(original_action_config)
with patch.dict(
MODALITY_CONFIGS[EMBODIMENT.value]["action"].__dict__,
{"action_configs": new_configs},
):
mutated = _compute_relative_action_fingerprint(EMBODIMENT, "eef_9d")
assert mutated != baseline
def test_changes_when_delta_indices_change(self):
baseline = _compute_relative_action_fingerprint(EMBODIMENT, "eef_9d")
action_modality = MODALITY_CONFIGS[EMBODIMENT.value]["action"]
with patch.dict(
action_modality.__dict__,
{"delta_indices": [0, 1, 2]},
):
mutated = _compute_relative_action_fingerprint(EMBODIMENT, "eef_9d")
assert mutated != baseline
# ---------------------------------------------------------------------------
# generate_rel_stats — cache hit / miss / mixed via mocked calculate_stats_for_key
# ---------------------------------------------------------------------------
class TestGenerateRelStatsCache:
def test_first_run_computes_and_persists_fingerprints(self, dataset_dir, mock_calculate):
generate_rel_stats(dataset_dir, EMBODIMENT)
assert sorted(c[2] for c in mock_calculate) == sorted(RELATIVE_KEYS)
with open(dataset_dir / LE_ROBOT_REL_STATS_FILENAME) as f:
payload = json.load(f)
assert STATS_FINGERPRINTS_KEY in payload
for key in RELATIVE_KEYS:
assert key in payload
assert key in payload[STATS_FINGERPRINTS_KEY]
assert payload[STATS_FINGERPRINTS_KEY][key].startswith("sha256:")
def test_second_run_is_full_cache_hit(self, dataset_dir, mock_calculate):
generate_rel_stats(dataset_dir, EMBODIMENT)
mock_calculate.clear()
generate_rel_stats(dataset_dir, EMBODIMENT)
assert mock_calculate == [], "fresh fingerprints must produce zero recompute"
def test_legacy_file_without_fingerprints_is_regenerated(self, dataset_dir, mock_calculate):
"""Pre-existing relative_stats.json from before this fix must be recomputed."""
legacy_payload = {key: {"_legacy": True} for key in RELATIVE_KEYS}
with open(dataset_dir / LE_ROBOT_REL_STATS_FILENAME, "w") as f:
json.dump(legacy_payload, f)
generate_rel_stats(dataset_dir, EMBODIMENT)
assert sorted(c[2] for c in mock_calculate) == sorted(RELATIVE_KEYS)
with open(dataset_dir / LE_ROBOT_REL_STATS_FILENAME) as f:
payload = json.load(f)
for key in RELATIVE_KEYS:
assert "_legacy" not in payload[key]
assert payload[STATS_FINGERPRINTS_KEY][key].startswith("sha256:")
def test_stale_fingerprint_triggers_regenerate(self, dataset_dir, mock_calculate):
generate_rel_stats(dataset_dir, EMBODIMENT)
with open(dataset_dir / LE_ROBOT_REL_STATS_FILENAME) as f:
payload = json.load(f)
payload[STATS_FINGERPRINTS_KEY]["eef_9d"] = "sha256:" + "0" * 64
with open(dataset_dir / LE_ROBOT_REL_STATS_FILENAME, "w") as f:
json.dump(payload, f)
mock_calculate.clear()
generate_rel_stats(dataset_dir, EMBODIMENT)
recomputed = [c[2] for c in mock_calculate]
assert recomputed == ["eef_9d"], f"only the stale key should regenerate; got {recomputed}"
def test_modality_config_change_invalidates_cache(self, dataset_dir, mock_calculate):
"""The whole point of the fingerprint: silently changed config must not get a stale cache hit."""
generate_rel_stats(dataset_dir, EMBODIMENT)
mock_calculate.clear()
action_modality = MODALITY_CONFIGS[EMBODIMENT.value]["action"]
idx = action_modality.modality_keys.index("eef_9d")
new_configs = list(action_modality.action_configs)
new_configs[idx] = replace(new_configs[idx], format=ActionFormat.XYZ_ROTVEC)
with patch.dict(action_modality.__dict__, {"action_configs": new_configs}):
generate_rel_stats(dataset_dir, EMBODIMENT)
recomputed = [c[2] for c in mock_calculate]
assert "eef_9d" in recomputed, "format change must trigger eef_9d regeneration"
assert "joint_position" not in recomputed, (
"joint_position config did not change; must not be regenerated"
)
def test_partial_cache_only_recomputes_missing(self, dataset_dir, mock_calculate):
"""Pre-fill cache for one key only; the other should be the only one computed."""
eef_fp = _compute_relative_action_fingerprint(EMBODIMENT, "eef_9d")
prefilled = {
"eef_9d": {k: v.tolist() for k, v in _stub_stats().items()},
STATS_FINGERPRINTS_KEY: {"eef_9d": eef_fp},
}
with open(dataset_dir / LE_ROBOT_REL_STATS_FILENAME, "w") as f:
json.dump(prefilled, f)
generate_rel_stats(dataset_dir, EMBODIMENT)
assert [c[2] for c in mock_calculate] == ["joint_position"]
with open(dataset_dir / LE_ROBOT_REL_STATS_FILENAME) as f:
payload = json.load(f)
for key in RELATIVE_KEYS:
assert key in payload
assert key in payload[STATS_FINGERPRINTS_KEY]
# ---------------------------------------------------------------------------
# Backward compat — downstream consumers must not see __fingerprints__ as data
# ---------------------------------------------------------------------------
def test_fingerprints_key_does_not_collide_with_action_keys():
"""Reserved sentinel must not match any registered action key string anywhere."""
for tag, cfg in MODALITY_CONFIGS.items():
action = cfg.get("action")
if action is None or action.action_configs is None:
continue
assert STATS_FINGERPRINTS_KEY not in action.modality_keys, (
f"reserved sentinel {STATS_FINGERPRINTS_KEY!r} collides with an "
f"action key in embodiment {tag!r}"
)
def test_per_action_key_payload_unchanged_shape(dataset_dir, mock_calculate):
"""The per-action_key dict written to disk must contain only the stat fields,
not the fingerprint — downstream loader iterates ``stats[key].keys()``
expecting ``{mean, std, min, max, q01, q99}``.
"""
generate_rel_stats(dataset_dir, EMBODIMENT)
with open(dataset_dir / LE_ROBOT_REL_STATS_FILENAME) as f:
payload = json.load(f)
for key in RELATIVE_KEYS:
assert set(payload[key].keys()) == {"mean", "std", "min", "max", "q01", "q99"}
|